uds-lsv / bert-lnl Goto Github PK

Hi, I am using your framework to train BERT under noisy labels. To see baseline performance, I am training BertWN on the IMDB dataset. I found that the accuracies don't change even a little bit. What happened? Could you help me?

Here is how I print them: (in bert_wn_trainer.py)

Here are the hyperparameters:

hi,thanks for your impressive work， when i reproduce it，i found some error or sonmething you did not mentioned，could you please help me, many thanks !

my data path is ./data/Ag_News/txt_data

Hello authors,
I found your paper very impressive and fascinating.

I was hoping if I could try some of your experiments.
Do you have specific plans on when to upload the code you used in the paper?

Thanks for the great work.